codex-remote-control – Use Claude's Remote Control with Codex Sessions

What happened

The HTML is raw GitHub page markup. Let me fetch the README directly to get the actual project facts. I now have the full facts from the README. Here is the article:

A developer going by sahrizvi on GitHub has published codex-remote-control, a Claude Code skill that bridges a long-standing gap between two popular AI coding tools: Codex CLI and Claude Code. The project surfaces on Hacker News at https://news.ycombinator.com/item?id=49526313 with one point and no comments at the time of this writing. The repository lives at https://github.com/sahrizvi/codex-remote-control.

This piece explains what codex-remote-control does, how the plumbing works, and who stands to benefit. It is aimed at developers who already run Codex CLI sessions for long-running coding tasks and want oversight of those sessions from anywhere, not just the terminal where they started.

How it works

Codex CLI has no built-in remote control. Once you start a long task and walk away from that terminal you lose visibility entirely: you cannot check on progress, you cannot redirect the agent, and you cannot tell whether it is working, idle, or silently blocked on a rate limit. Claude Code, by contrast, ships its own remote control feature and a companion phone app. The codex-remote-control project is a Claude Code skill that chains the two tools together, giving a running Claude session eyes and hands on any Codex CLI session running on the same machine.

The project ships as a single directory installed into the Claude skills folder. Installation from the marketplace requires two commands: one to add the repository as a marketplace source and one to install the skill. A manual path exists for developers who prefer not to register a new marketplace: clone the repository, then copy the skills subdirectory into either the global skills folder under the home directory or into a project-level skills folder. Either way the result is exactly one new directory added and no other skills touched.

The mechanism relies entirely on Claude Code acting as an intermediary. When you open Claude Code's remote control on your phone and ask what Codex is doing, Claude reads the Codex session's own state: the last message Codex produced in its own words, the commands it ran, and whether those commands exited successfully. That state is surfaced back to you as a conversation reply. The skill does not intercept Codex's network traffic, parse log files, or build a parallel monitoring daemon.

Why it matters

The other direction works the same way. When you type an instruction into the Claude Code phone app, the skill queues the instruction to the targeted Codex session. Codex picks it up the next time it reads input. The README demonstrates three distinct capabilities: checking what a session is doing and which commands it ran, sending new instructions mid-run without going to the desk, and distinguishing between an actively working session, an idle session that finished its turn, and a session that hit a rate-limit window and silently stopped. The third case, the README notes, is the reason the project exists.

The problem that codex-remote-control solves is an asymmetry between two tools that many developers now use together. Codex CLI is built for long autonomous coding runs that outlast a single sitting. Claude Code ships a phone app specifically for monitoring and steering agents remotely. Without the skill, a developer running Codex CLI has no path to those remote control capabilities at all. With it, the phone app becomes a window into Codex sessions without any configuration beyond the install.

Who is affected

The rate-limit visibility feature is the most operationally meaningful part of the announcement. A Codex session that has consumed its entire five-hour request window and stopped looks identical from outside the terminal to a session that is simply thinking. The only difference is that one will resume on its own and the other will not until the window resets. Knowing the difference without sitting at the terminal has direct value for anyone running overnight or commute-length coding tasks.

The immediate audience is any developer who uses both Codex CLI and Claude Code and who has experienced the frustration of not knowing whether a background session is making progress. Developers who run multiple sessions in parallel, for example one per git worktree or one per review branch, are particularly well served: the skill can list all running sessions, report their last active timestamp, and distinguish which are working and which have gone idle. The README shows a two-session example: one active in a project directory one minute ago, one idle in a temporary review worktree for two hours.

Developers who run only one of the two tools, or who do not use Claude Code's phone app, get nothing from this skill. There is also no Windows or Linux desktop UI path described in the repository; the remote control feature referenced is the Claude Code phone client. Teams operating in environments that restrict third-party Claude plugins should verify whether adding a marketplace source is permitted before attempting installation.

What to watch next

Because the skill ships through the Claude plugin marketplace, its maintenance and compatibility depend on Claude Code continuing to support that plugin architecture and the remote control API surface the skill uses. Builders adding this to a workflow should verify that the skill's session introspection still matches actual Codex CLI session behavior after any Codex CLI update, since the README describes reading Codex's own last message and exit codes rather than a versioned structured API. The update command listed in the README is claude plugin update codex-remote-control, and running it after either tool updates is the first verification step.

The repository had one Hacker News point and no comments at submission, which means community-reported edge cases, compatibility failures, or feature requests have not yet accumulated. Developers who adopt codex-remote-control early should pay attention to the GitHub issues page for the first reports of sessions where idle, stuck, and rate-limited are not correctly distinguished, since that three-way classification is the core claim of the skill. The source is at https://github.com/sahrizvi/codex-remote-control.